HTTP is the main protocol Gatling targets, so that’s where we place most of our effort.
Gatling HTTP allows you to load test web applications, web services or websites. It supports HTTP and HTTPS with almost every existing feature of common browsers such as caching, cookies, redirect, etc.
http object in order to create an HTTP protocol.
As every protocol in Gatling, the HTTP protocol can be configured for a scenario. This is done thanks to the following statements:
val httpProtocol = http.baseUrl("http://my.website.tld") val scn = scenario("myScenario") // etc... setUp(scn.inject(atOnceUsers(1)).protocols(httpProtocol))
As you may have seen in the previous example, you can set a base URL.
This base URL will be prepended to all urls that does not start with
val httpProtocol = http.baseUrl("http://my.website.tld") val scn = scenario("My Scenario") .exec( http("My Request") .get("/my_path") ) // will make a request to "http://my.website.tld/my_path" .exec( http("My Other Request") .get("http://other.website.tld") ) // will make a request to "http://other.website.tld" setUp(scn.inject(atOnceUsers(1)).protocols(httpProtocol))
Load testing several servers with client based load balancing¶
If you want to load test several servers at the same time, to bypass a load-balancer for example, you can use methods named
baseUrls which accepts a
String* or a
val httpProtocol = http.baseUrls("http://my1.website.tld", "http://my2.website.tld", "http://my3.website.tld")
The selection of the URL is made once and for all for a given virtual user.
Automatic warm up¶
The Java/NIO engine start up introduces an overhead on the first request to be executed. In order to compensate this effect, Gatling automatically performs a request to https://gatling.io.
To disable this feature, just add
.disableWarmUp to an HTTP Protocol Configuration definition.
To change the warm up url, just add
// override warm up URL to http://www.google.com val httpProtocol = http.warmUp("http://www.google.com") // disable warm up val httpProtocolNoWarmUp = http.disableWarmUp
Max connection per host¶
In order to mimic real web browser, Gatling can run multiple concurrent connections per virtual user when fetching resources on the same hosts.
By default, Gatling caps the number of concurrent connections per remote host per virtual user to 6, but you can change this number with
Gatling ships a bunch of built-ins for well-known browsers:
// 10 connections per host. val httpProtocolMax10Connections = http.maxConnectionsPerHost(10) // Firefox max connections per host preset. val httpProtocolMaxConnectionsLikeFirefox = http.maxConnectionsPerHostLikeFirefox
In Gatling 1, connections are shared amongst users until 1.5 version. This behavior does not match real browsers, and doesn’t support SSL session tracking.
In Gatling 2, the default behavior is that every user has his own connection pool.
This can be tuned with the
HTTP/2 experimental support can be enabled with the .enableHttp2 option.
Note that you’ll either need your injectors to run with Java 9+, or make sure that gatling.http.ahc.useOpenSsl wasn’t turned to false in Gatling configuration.
val httpProtocol = http.enableHttp2
HTTP/2 Push is currently not supported.
When HTTP/2 is enabled Gatling will try to connect to your remotes using HTTP/2 through the ALPN protocol. If your remote supports HTTP/2, Gatling will use the protocol, and fall back to HTTP/1 otherwise. There is no specific code to add in the middle of your requests.
Next time you use that remote with the same user, if Gatling knows that your remote doesn’t support HTTP/2, it won’t try again and therefore won’t use ALPN.
One of the main purpose of HTTP/2 is to support multiplexing. This means that on a single connection, you are able to send multiple requests, without waiting for the responses,
and receive these responses in whatever order.
It means that, using HTTP/2, browsers and Gatling won’t open additional connections to the same remote for a given virtual user (assuming you don’t enable shareConnections`) once they know that the remote is using HTTP/2.
The first time Gatling encounters a remote, the connections will be opened like in HTTP/1 mode if there are multiple requests (for example in a
If the remote is using HTTP/1, these connections will be used if needed. If it is using HTTP/2, a single connection will be maintained, and the other ones will reach idle timeout and be closed.
It is possible to populate the Gatling cache concerning protocol and remotes before the run, using the
http2PriorKnowledgeMap(Map[String, Boolean]) method on the protocol.
With this method, you are able to tell Gatling that remotes support HTTP/2 or not. It means that if you are setting a remote to true (it supports HTTP/2), additional connections won’t be created the first time the remote is encountered in the simulation. If you are setting a remote to false (it doesn’t support HTTP/2), ALPN won’t be used, and additional connections will be created.
This option is useful to simulate users that already went to your website, and whose browsers already cached the fact that your website is using HTTP/2 or HTTP/1.
- If you configure a remote in prior knowledge and set it to true, but that the ALPN ends in the remote only supporting HTTP/1, the request will crash.
- Use the
http2PriorKnowledgeoption only if you are sure about your remote configuration.
DNS Name Resolution¶
By default, Gatling uses Java’s DNS name resolution, meaning that it uses a cache shared amongst all virtual users.
More over, Java’s cache doesn’t honor DNS records TTL.
You can control the TTL with
-Dsun.net.inetaddr.ttl=N where N is a number of seconds.
If you’re using the JDK resolution and have multiple IP (multiple DNS records) for a given hostname, Gatling will automatically shuffle them to emulate DNS round-robin.
You can use a Netty based DNS resolution instead, with
This method can take a sequence of DNS server adresses, eg
If you don’t pass DNS servers, Gatling will use the one from your OS configuration on Linux and MacOS only,
and to Google’s ones on Windows(don’t run with heavy load as Google will block you).
You can also make it so that every virtual user performs its own DNS name resolution with
This parameter is only effective when using
Note this feature is experimental. This feature is pretty useful if you’re dealing with an elastic cluster where new IPs are added to the DNS server under load, for example with AWS ALB and Route53.
You can of course define hostname aliases at the OS level in the
But you can also pass a
Map[String, String] to
.hostNameAliases where values are valid IP addresses.
Note that, just like with
/etc/hosts you can only define one IP per alias.
One can set a different Host than the url one:
You can bind the sockets from specific local addresses instead of the default one:
localAddress(localAddress: String) localAddresses(localAddress1: String, localAddress2: String)
When setting multiple addresses, each virtual user is assigned to one single local address once and for all.
By default, Gatling uses the KeyManagerFactory configuration defined in gatling.conf, or if undefined, falls back to the JVM’s default one.
Then, it’s possible to have per virtual user KeyManagerFactories, typically if you want them to use different sets of keys:
perUserKeyManagerFactory(f: Long => KeyManagerFactory)
This function’s input is the virtual user’s id (if you need it to generate some file name) and returns a javax.net.ssl.KeyManagerFactory.
Request building parameters¶
Referer HTTP header can be automatically computed.
This feature is enabled by default.
To disable this feature, just add
.disableAutoReferer to an HTTP Protocol Configuration definition.
Gatling caches responses using :
- Expires header
- Cache-Control header
- Last-Modified header
To disable this feature, just add
.disableCaching to an HTTP Protocol Configuration definition.
When a response gets cached, checks are disabled.
Url components are supposed to be urlencoded. Gatling will encode them for you, there might be some corner cases where already encoded components might be encoded twice.
If you know that your urls are already properly encoded, you can disable this feature with
Note that this feature can also be disabled per request.
Request stats are logged and then used to produce reports. Sometimes, some requests may be important for you for generating load, but you don’t actually want to report them. Typically, reporting all static resources might generate a lot of noise, and yet failed static resources might not be blocking from a user experience perspective.
Gatling provides several means to turn requests silent.
Silent requests won’t be reported and won’t influence error triggers such as tryMax and exitHereIfFailed.
Yet, response times will be accounted for in
Some parameters are available here at protocol level, some others are available at request level.
- explicitly turning a given request silent or notSilent has precedence over everything else
- otherwise, a request is silent if it matches protocol’s
- otherwise, a request is silent if it’s a resource (not a top level request) and protocol’s
silentResourcesflag has been turned on
- otherwise, a request is not silent
silentUri lets you pass a regular expression that would disable logging for ALL matching requests:
silentResources silences all resource requests, except the ones that were explicitly turned
Gatling lets you set some generic headers at the http protocol definition level with:
header(name: String, value: Expression[String]): set a single header.
headers(headers: Map[String, String]): set a bunch of headers.
.header("foo", "bar") .headers(Map("foo" -> "bar", "baz" -> "qix"))
You have also the following built-ins for the more commons headers:
acceptHeader(value: Expression[String]): set
acceptCharsetHeader(value: Expression[String]): set
acceptEncodingHeader(value: Expression[String]): set
acceptLanguageHeader(value: Expression[String]): set
authorizationHeader(value: Expression[String]): set
connectionHeader(value: Expression[String]): set
contentTypeHeader(value: Expression[String]): set
doNotTrackHeader(value: Expression[String]): set
userAgentHeader(value: Expression[String]): set
You can set a function to sign a request once Gatling has built it, just before it’s being sent over the wire:
We also provide a built-in for OAuth1:
signWithOAuth1(consumerKey: Expression[String], clientSharedSecret: Expression[String], token: Expression[String], tokenSecret: Expression[String])
Response handling parameters¶
By default Gatling automatically follow redirects in case of 301, 302, 303, 307 or 308 response status code, you can disable this behavior with
To avoid infinite redirection loops, Gatling sets a limit on the number of redirects.
The default value is 20. You can tune this limit with:
By default, Gatling will change the method to “GET” on 302 to conform to most user agents’ behavior.
You can disable this behavior with
Some people might want to process manually the response. Gatling protocol provides a hook for that need:
You can define checks at the http protocol definition level with:
They will be apply on all the requests, however you can disable them for given request thanks to the
Gatling can fetch resources in parallel in order to emulate the behavior of a real web browser.
At the protocol level, you can use
inferHtmlResources methods, so Gatling will automatically parse HTML to find embedded resources and load them asynchronously.
The supported resources are:
- import directives in HTML and @import CSS rule.
You can also specify black/white list or custom filters to have a more fine grain control on resource fetching.
BlackList take a sequence of pattern, eg
Seq("http://www.google.com/.*", "http://www.github.com/.*"), to include and exclude respectively.
inferHtmlResources(white: WhiteList): fetch all resources matching a pattern in the white list.
inferHtmlResources(white: WhiteList, black: BlackList): fetch all resources matching a pattern in the white list excepting those in the black list.
inferHtmlResources(black: BlackList, white: WhiteList = WhiteList(Nil)): fetch all resources excepting those matching a pattern in the black list and not in the white list.
Finally, you can specify the strategy for naming those requests in the reports:
nameInferredHtmlResourcesAfterUrlTail``(default): name requests after the resource's url tail (after last ``/)
nameInferredHtmlResourcesAfterPath: name requests after the resource’s path
nameInferredHtmlResourcesAfterAbsoluteUrl: name requests after the resource’s absolute url
nameInferredHtmlResourcesAfterRelativeUrl: name requests after the resource’s relative url
nameInferredHtmlResourcesAfterLastPathElement: name requests after the resource’s last path element
nameInferredHtmlResources(f: Uri => String): name requests with a custom strategy
You can tell Gatling to use a proxy to send the HTTP requests. You can optionally set a different port for HTTPS and credentials:
val httpProtocol = http .proxy( Proxy("myHttpProxyHost", 8080) .httpsPort(8143) .credentials("myUsername", "myPassword") ).proxy( Proxy("mySocks4ProxyHost", 8080) .socks4 ).proxy( Proxy("mySocks5ProxyHost", 8080) .httpsPort(8143) .socks5 )
You can also disable the use of proxy for a given list of hosts with
val httpProtocol = http .proxy(Proxy("myProxyHost", 8080)) .noProxyFor("www.github.com", "www.akka.io")